Catálogo de publicaciones - libros

Image quality assessment plays an important role in several image processing applications, including data approximation by triangular meshes. The determination of adequate metrics is essential for constructing algorithms that generate high quality models. This paper evaluates a number of different image measures used to refine a given triangular mesh until a specified accuracy is obtained, which are more effective than traditional metrics such as the magnitude of the maximum vertical distance between pairs of corresponding points in the images. Experiments show that a considerable reduction in the triangulation size can be obtained by using more effective criteria for selecting the data points. Several metrics for evaluating the overall quality of the resulting models are also presented and compared.

Grassmann-Cayley algebra (GCA) provides an efficient formulation of projectice geometry, allowing work with elementary geometrical objects at a low computational cost. In this paper we use GCA as a mathematical framework for modeling conic curves and quadric surfaces of 3D space, and for computing rational parameterizations of these. Then, through ad hoc sampling of the parameter space, we derive tessellations of conics and quadrics.

A person tracking algorithm in video scenes shot by camera in motion is proposed. It is shown experimentally that for such a recording scenario an uncomplicated algorithm incorporating background motion subtraction can successfully track moving human silhouettes.

The problems of constructing the points sequences with fast uniform distribution in the fractal domains associated with canonical number systems in quadratic fields are considered.

This paper focuses on the problem of vision-based tracking of multiple objects. Probabilistic tracking in 3D supported by multiple video streams allows us to formalize an efficient observation model that is robust to occlusions. Each tracked object is assigned a , a probabilistically meaningful pixel occupancy map, supplying weights used in the calculation of other objects observation likelihood. A implementation demonstrates the robustness of the resulting tracking system on synthetic data.

The aim of this paper is to propose a system for tracking of facial features. Face blob will be segmented using stochastic skin-color model. The features will be looked for in the found face blob. Thereafter we detect eyebrows, eyes, and nostrils in the area of the face. The system is only applicable under the constant lighting conditions. The results seem promising.

In this paper we propose an efficient approach for Motion Vector (MV) based object detection in MPEG-1 video streams. Based on experimental results performed over the MPEG testing dataset and measuring performance using the standard recall and precision metrics for perceptual performance, and run time metric for efficiency performance. Our approach is remarkably superior to the alternative techniques. In addition to these results, we describe a user system interface that we developed, where users can maintain the configurations interactively.

The work deals with possibilities of approximation of non-square areas using combined discrete orthogonal transforms (DOT), where image data are represented using less spectral coefficients than in the case of single transform approach. The work is based on an assumption that properly chosen orthogonal transforms complete each other in the process of image approximation.

Recently, a new kind of images taken by a camera in a “real-world” environment appeared. It implies different strong degradations missing in scanner-based pictures and the presence of complex backgrounds. In order to segment text more properly as possible, a new binarization technique is proposed using color information. This information is used at proper moments in the processing not from the beginning to have smaller regions-of-interest. It presents the advantage of reducing the computation time. In this paper, an accent is put on stroke analysis and character segmentation. The binarization method takes it into account in order to improve character segmentation and recognition afterwards.

In transport security context, real-time detection of potentially dangerous situations is very important. We need to be able to identify pedestrians like intruders in forbidden areas, as quickly as possible. The signatures must be precise in order to identify persons, and also compact to be rapidely transmitted through a network. We propose in this paper a new signature which is based on a Principal Component Analysis (PCA), applied to color histograms.